High-level sport accepts, in silence, a failure rate that would be intolerable in any other control domain: its primary doping detection tool lets between 95 and 98 dopers out of 100 slip through. This is not an activist estimate. It is the conclusion of a collective expert assessment by Inserm published in April 2026, based on the analysis of more than 3,800 scientific publications.

The athlete’s biological passport, presented since its introduction in 2008 as a turning point in anti-doping efforts, detects less than 5% of actual doping cases. In the most affected disciplines, the estimated prevalence of doping reaches 30%. Controls capture only an infinitesimal fraction of it.

The Essentials

  • The Inserm collective expert assessment (April 2026), based on more than 3,800 publications, concludes that the biological passport detects less than 5% of actual doping cases.
  • The estimated prevalence of doping reaches up to 30% in certain disciplines, compared with less than 1% of official positive test rates.
  • This result is the consequence of a deliberate choice: a reliability threshold set at 99.9% to protect the innocent, which renders the system nearly blind to the guilty.
  • Cycling has been experimenting since June 2026 with a power passport, based on performance rather than blood; AI opens a third path by re-reading existing biological data differently.
  • The central dilemma is as much political as statistical: lowering the proof threshold would catch more guilty parties at the risk of condemning the innocent.

30% Dopers, Less Than 1% Positives: The Gap Between Estimation and Detection

Official doping figures are reassuring. The World Anti-Doping Agency (WADA) publishes annually a rate of positive tests—expressed as adverse analytical findings (AAF)—which stands at around 0.8% of tests conducted (0.77% in 2022, 0.80% in 2023). This figure is real. It does not measure doping; it measures what controls manage to detect.

Inserm compared these official statistics with available epidemiological data, particularly studies based on indirect reporting or anonymous methods, which circumvent the obvious bias of self-reporting. The result is brutal: according to disciplines and estimation methods, the true prevalence of doping lies between 14% and 30% of elite athletes. In certain analyzed cohorts, estimates exceed 40%.

The gap between the less than 1% official figure and the estimated 30% is not a scandal strictly speaking. It is a mechanical consequence of how the biological passport functions.

The passport does not search for prohibited substances in urine or blood. It tracks over time a series of biomarkers specific to each athlete, primarily hematological parameters such as hemoglobin levels or the percentage of reticulocytes. The idea is elegant: if an athlete’s values deviate suspiciously from their individual profile, an anomaly is flagged. The approach circumvents undetectable substances and new molecules that direct tests do not yet know how to identify.

The problem lies in the calibration. For an anomaly to trigger a procedure, it must be statistically incompatible with natural variations at a confidence level of 99.9%. This threshold was set to avoid any unjust conviction. A single false positive—an innocent athlete sanctioned—would represent irreparable damage to their career and to the system’s credibility. This choice is ethically defensible.

Statistically, it renders the passport nearly blind. Modern doping products, particularly microdoses of erythropoietin (EPO) and peptide products, are precisely designed to produce significant physiological effects without triggering the crude variations the passport monitors. Contemporary doping has adapted to the detection tool.

A Tool Calibrated to Protect the Innocent, at the Price of Near-Total Impunity for the Guilty

This dilemma has a name in statistical theory: it opposes Type I error (convicting an innocent person) to Type II error (acquitting a guilty person). In most legal systems, the priority accorded to protecting the innocent is the very foundation of the presumption of innocence. “Better to let ten guilty people go free than to condemn one innocent person” is a principle rooted in liberal societies.

Applied to sports doping, this principle produces a paradoxical result: a system that almost perfectly guarantees the integrity of the innocent while almost equally guaranteeing the impunity of the guilty. The 99.9% threshold is not a design flaw; it is an ethical choice with considerable practical consequences.

The Inserm report points to another structural limit: the detection window. Most hematological biomarkers return to normal values within days to weeks after treatment stops. An athlete who times their use far from competitions and phases of intensive testing mechanically escapes monitoring. Out-of-competition testing has been reinforced precisely to reduce this window, but locating athletes remains an operational constraint.

There exists a third limit, less discussed: interindividual variability. Some athletes naturally present atypical biological profiles, with elevated hematological values linked to altitude training, genetic factors, or benign pathologies. The passport must distinguish natural atypicality from chemical atypicality. In ambiguous cases, statistical caution demands no conclusion be drawn, which amounts to classifying the anomaly as inconclusive.

Cycling Tests a Power Passport Since June 2026

Faced with these limits, two paths are emerging. The first is institutional and a decade old: improve the existing biological passport by refining its parameters and integrating new biomarkers. The second is more radical: change the field of surveillance.

Professional cycling has taken this second step. Since June 2026, the Union Cycliste Internationale (UCI) has been experimenting with a power passport in a pilot group. The principle differs from its hematological predecessor: rather than monitoring blood, the power passport tracks physical performance measured in watts per kilogram of body mass, in relation to physiological parameters such as heart rate and oxygen consumption.

The idea rests on a simple observation: an athlete who dopes improves their ability to produce power at a given effort level. This improvement leaves a trace in their performance data, even if their blood profile remains within acceptable limits. Power sensors have equipped all professional bikes for about a decade; the volume of available data is considerable.

The power passport presents its own advantages and difficulties. It is harder to manipulate chemically, because the effects of a doping product on performance are less predictable than its effects on a specific biomarker. Conversely, performance variations are influenced by numerous and legitimate factors: fitness state, weather conditions, race strategy, accumulated fatigue. Isolating the doping component in this noise is a statistical challenge of a different nature than the blood passport.

The UCI has not yet published results on this experimental phase. The stated objective is not to replace the biological passport, but to complement it. An athlete presenting anomalies in both passports simultaneously would provide a more robust body of evidence than an anomaly isolated in one or the other.

What AI Changes in Reading Biological Data

The second path is algorithmic. Research teams, notably within European laboratories working with WADA, are exploring the application of artificial intelligence to reading existing biological passport data.

The classical approach to the passport rests on parametric statistical models: one assumes that biomarkers follow certain known distributions, and one measures an individual’s deviation from these distributions. Machine learning models can, in theory, detect subtle patterns in the combination of multiple biomarkers that individual analyses do not capture. A blood profile may trigger no alert on any parameter taken in isolation, but reveal a statistically abnormal configuration when examining the correlation between ten parameters simultaneously.

Preliminary results suggest that these approaches could improve detection sensitivity without degrading specificity, that is, without increasing the number of false positives. But these works remain in validation phases. Applying an AI model to decisions that engage an athlete’s career requires a level of validation and explainability that current algorithms do not yet fully satisfy.

The question of algorithmic transparency is central. An athlete sanctioned based on an opaque machine learning model would have legitimate reasons to contest the decision before the Court of Arbitration for Sport. AI can refine detection; it does not resolve the burden of proof problem.

It is worth noting that this challenge of algorithmic transparency arises in other domains where high-stakes individual decisions rest on predictive models, whether in credit, recruitment, or medicine. Computing power is becoming the scarce resource of the century, but the scarce resource in anti-doping efforts is not processing capacity: it is the ability to transform a statistical correlation into legally actionable proof.

Inserm Recommends a Framework for Action on Four Axes

The Inserm collective expert assessment is not limited to diagnosis. It formulates recommendations articulated around four axes.

The first is research on biomarkers. Inserm advocates investing in identifying new biological markers, particularly genomic and proteomic markers, which could detect the effects of substances not yet identified or emerging doping approaches such as gene therapy. This path is promising but on a horizon of several years.

The second axis is the revision of decision thresholds. Inserm opens the question without resolving it: is it possible to differentiate the threshold according to context, by applying different proof requirements depending on whether one seeks to sanction an athlete or to trigger a thorough investigation? A signal at 98% confidence is not sufficient to condemn, but could be sufficient to intensify targeted testing on an individual. This graduation of the threshold would represent a significant change in doctrine.

The third axis concerns international coordination. The multiplicity of national anti-doping agencies, with their varying practices, resources, and independence, limits the coherence of the global system. Inserm points out that some national federations lack the means to rigorously apply the WADA framework. The funding question is not trivial: WADA has an annual budget of approximately 44 million dollars to monitor global sport, a modest sum relative to the task.

The fourth axis is preventive and educational. Inserm recalls that deterrence through punishment is not the only available lever. Targeted prevention programs for junior athletes, coaches, and medical entourages could reduce entry into doping, independent of testing effectiveness. This is a public health logic applied to sport.

Is the Estimated 30% Prevalence Reliable?

One element deserves to be addressed honestly: the prevalence estimates on which the diagnosis rests are themselves uncertain.

Studies concluding to 14-30% doping rely on indirect methods: the randomised response technique, which allows athletes to answer sensitive questions anonymously via a probabilistic protocol, or prevalence studies based on biomarkers without individual evidentiary value but revealing at a population scale. These methods are methodologically sound; they do not produce absolute certainties.

It is possible that the true prevalence is lower than the high estimates. It is also possible that it is higher in certain contexts that the studies did not cover. What Inserm says with precision is that the gap between the less than 1% of official positives and epidemiological estimates is too wide to be explained by methodological biases. The order of magnitude of under-detection is robust, even if the exact figure of “less than 5%” carries a margin of uncertainty.

This distinction matters. Inserm does not say that 95% of athletes are doped. It says that among athletes who dope themselves, the biological passport identifies fewer than one in twenty. This is not the same thing, and precision is necessary to avoid unduly discrediting elite sport as a whole.

Professional football, regularly cited in the report alongside cycling, illustrates this point. Available prevalence studies are rarer and more heterogeneous for this sport than for cycling or athletics, where the culture of measurable individual performance facilitates epidemiological research. Inserm’s conclusions apply with varying degrees of certainty depending on the disciplines.

Anti-Doping Efforts Face a Choice About Sports Civilization

The Inserm report poses implicitly a question that sports bodies have not yet formulated publicly: what level of detection is it reasonable to demand, and at what cost?

Lowering the biological passport threshold from 99.9% to 99% would substantially improve test sensitivity, but would significantly multiply the risk of false positives. For considerable testing volumes, this represents potentially a significant number of careers destroyed wrongly. For WADA and federations, this choice is politically impractical.

Maintaining the current threshold, conversely, amounts to accepting that anti-doping efforts have essentially a symbolic and deterrent function, not a systematic detection function. They catch the careless, the unlucky, and a few whose medical entourages miscalculated their dosing. They let through the professionals of planned doping.

The power passport and AI do not resolve this fundamental dilemma; they displace it. A more sensitive multiparametric surveillance system does not change the threshold problem: it poses it in a more complex space, where false positives are harder to anticipate and contest.

What the cycling experiments and Inserm’s work allow us to hope for is a progressive convergence: more sensitive tools, a threshold doctrine graduated according to the stakes of the decision, and reinforced international coordination. Criminal investigations, in which insufficient evidence for conviction can justify increased surveillance, offer a partial model. An athlete whose two passports sound simultaneously, without each one triggering sanction by itself, could be subject to reinforced testing that would produce actionable evidence.

This is a different logic from the search for biological flagrante delicto. And it presupposes governance of professional sport more transparent about its own limits than that which has prevailed for twenty years.

Sources

  1. Curieux.live — “Footballeurs et cyclistes pros : une étude scientifique démontre que le dopage est bien plus important que mesuré” (July 2026): https://www.curieux.live/2026/07/03/footballeurs-et-cyclistes-pros-une-etude-scientifique-demontre-que-le-dopage-est-bien-plus-important-que-mesure/
  2. Inserm — Collective expert assessment on doping in elite sport, April 2026 (full report available on inserm.fr)
  3. World Anti-Doping Agency (WADA) — Annual statistical report on anti-doping testing, wada-ama.org
  4. Union Cycliste Internationale (UCI) — Press release on power passport experimentation, June 2026, uci.org
  5. Inserm – Official Collective Expert Assessment Doping (April 2026)
  6. Inserm – Official Press Room (April 24, 2026)
  7. WADA – Anti-Doping Testing Figures 2023
  8. WADA – Athlete Biological Passport (official page)
  9. PMC – Confounding Factors of the Biological Passport (Springer 2021)
  10. Sports Integrity Initiative – Hon et al. 2015 Study on Prevalence
  11. Cyclingnews – Study ITA Power Passport (June 2026)
  12. WADA – Research EPO Micro-dosing